]>
Commit | Line | Data |
---|---|---|
de6cc651 | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 LT |
2 | /* |
3 | * Copyright (C) 2001 Momchil Velikov | |
4 | * Portions Copyright (C) 2001 Christoph Hellwig | |
cde53535 | 5 | * Copyright (C) 2005 SGI, Christoph Lameter |
7cf9c2c7 | 6 | * Copyright (C) 2006 Nick Piggin |
78c1d784 | 7 | * Copyright (C) 2012 Konstantin Khlebnikov |
6b053b8e MW |
8 | * Copyright (C) 2016 Intel, Matthew Wilcox |
9 | * Copyright (C) 2016 Intel, Ross Zwisler | |
1da177e4 LT |
10 | */ |
11 | ||
0a835c4f MW |
12 | #include <linux/bitmap.h> |
13 | #include <linux/bitops.h> | |
460488c5 | 14 | #include <linux/bug.h> |
e157b555 | 15 | #include <linux/cpu.h> |
1da177e4 | 16 | #include <linux/errno.h> |
0a835c4f MW |
17 | #include <linux/export.h> |
18 | #include <linux/idr.h> | |
1da177e4 LT |
19 | #include <linux/init.h> |
20 | #include <linux/kernel.h> | |
0a835c4f | 21 | #include <linux/kmemleak.h> |
1da177e4 | 22 | #include <linux/percpu.h> |
0a835c4f MW |
23 | #include <linux/preempt.h> /* in_interrupt() */ |
24 | #include <linux/radix-tree.h> | |
25 | #include <linux/rcupdate.h> | |
1da177e4 | 26 | #include <linux/slab.h> |
1da177e4 | 27 | #include <linux/string.h> |
02c02bf1 | 28 | #include <linux/xarray.h> |
1da177e4 LT |
29 | |
30 | ||
1da177e4 LT |
31 | /* |
32 | * Radix tree node cache. | |
33 | */ | |
58d6ea30 | 34 | struct kmem_cache *radix_tree_node_cachep; |
1da177e4 | 35 | |
55368052 NP |
36 | /* |
37 | * The radix tree is variable-height, so an insert operation not only has | |
38 | * to build the branch to its corresponding item, it also has to build the | |
39 | * branch to existing items if the size has to be increased (by | |
40 | * radix_tree_extend). | |
41 | * | |
42 | * The worst case is a zero height tree with just a single item at index 0, | |
43 | * and then inserting an item at index ULONG_MAX. This requires 2 new branches | |
44 | * of RADIX_TREE_MAX_PATH size to be created, with only the root node shared. | |
45 | * Hence: | |
46 | */ | |
47 | #define RADIX_TREE_PRELOAD_SIZE (RADIX_TREE_MAX_PATH * 2 - 1) | |
48 | ||
0a835c4f MW |
49 | /* |
50 | * The IDR does not have to be as high as the radix tree since it uses | |
51 | * signed integers, not unsigned longs. | |
52 | */ | |
53 | #define IDR_INDEX_BITS (8 /* CHAR_BIT */ * sizeof(int) - 1) | |
54 | #define IDR_MAX_PATH (DIV_ROUND_UP(IDR_INDEX_BITS, \ | |
55 | RADIX_TREE_MAP_SHIFT)) | |
56 | #define IDR_PRELOAD_SIZE (IDR_MAX_PATH * 2 - 1) | |
57 | ||
7ad3d4d8 MW |
58 | /* |
59 | * The IDA is even shorter since it uses a bitmap at the last level. | |
60 | */ | |
61 | #define IDA_INDEX_BITS (8 * sizeof(int) - 1 - ilog2(IDA_BITMAP_BITS)) | |
62 | #define IDA_MAX_PATH (DIV_ROUND_UP(IDA_INDEX_BITS, \ | |
63 | RADIX_TREE_MAP_SHIFT)) | |
64 | #define IDA_PRELOAD_SIZE (IDA_MAX_PATH * 2 - 1) | |
65 | ||
1da177e4 LT |
66 | /* |
67 | * Per-cpu pool of preloaded nodes | |
68 | */ | |
69 | struct radix_tree_preload { | |
2fcd9005 | 70 | unsigned nr; |
1293d5c5 | 71 | /* nodes->parent points to next preallocated node */ |
9d2a8da0 | 72 | struct radix_tree_node *nodes; |
1da177e4 | 73 | }; |
8cef7d57 | 74 | static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, }; |
1da177e4 | 75 | |
148deab2 MW |
76 | static inline struct radix_tree_node *entry_to_node(void *ptr) |
77 | { | |
78 | return (void *)((unsigned long)ptr & ~RADIX_TREE_INTERNAL_NODE); | |
79 | } | |
80 | ||
a4db4dce | 81 | static inline void *node_to_entry(void *ptr) |
27d20fdd | 82 | { |
30ff46cc | 83 | return (void *)((unsigned long)ptr | RADIX_TREE_INTERNAL_NODE); |
27d20fdd NP |
84 | } |
85 | ||
02c02bf1 | 86 | #define RADIX_TREE_RETRY XA_RETRY_ENTRY |
db050f29 | 87 | |
d7b62727 MW |
88 | static inline unsigned long |
89 | get_slot_offset(const struct radix_tree_node *parent, void __rcu **slot) | |
db050f29 | 90 | { |
76f070b4 | 91 | return parent ? slot - parent->slots : 0; |
db050f29 MW |
92 | } |
93 | ||
35534c86 | 94 | static unsigned int radix_tree_descend(const struct radix_tree_node *parent, |
9e85d811 | 95 | struct radix_tree_node **nodep, unsigned long index) |
db050f29 | 96 | { |
9e85d811 | 97 | unsigned int offset = (index >> parent->shift) & RADIX_TREE_MAP_MASK; |
d7b62727 | 98 | void __rcu **entry = rcu_dereference_raw(parent->slots[offset]); |
db050f29 | 99 | |
db050f29 MW |
100 | *nodep = (void *)entry; |
101 | return offset; | |
102 | } | |
103 | ||
35534c86 | 104 | static inline gfp_t root_gfp_mask(const struct radix_tree_root *root) |
612d6c19 | 105 | { |
f8d5d0cc | 106 | return root->xa_flags & (__GFP_BITS_MASK & ~GFP_ZONEMASK); |
612d6c19 NP |
107 | } |
108 | ||
643b52b9 NP |
109 | static inline void tag_set(struct radix_tree_node *node, unsigned int tag, |
110 | int offset) | |
111 | { | |
112 | __set_bit(offset, node->tags[tag]); | |
113 | } | |
114 | ||
115 | static inline void tag_clear(struct radix_tree_node *node, unsigned int tag, | |
116 | int offset) | |
117 | { | |
118 | __clear_bit(offset, node->tags[tag]); | |
119 | } | |
120 | ||
35534c86 | 121 | static inline int tag_get(const struct radix_tree_node *node, unsigned int tag, |
643b52b9 NP |
122 | int offset) |
123 | { | |
124 | return test_bit(offset, node->tags[tag]); | |
125 | } | |
126 | ||
35534c86 | 127 | static inline void root_tag_set(struct radix_tree_root *root, unsigned tag) |
643b52b9 | 128 | { |
f8d5d0cc | 129 | root->xa_flags |= (__force gfp_t)(1 << (tag + ROOT_TAG_SHIFT)); |
643b52b9 NP |
130 | } |
131 | ||
2fcd9005 | 132 | static inline void root_tag_clear(struct radix_tree_root *root, unsigned tag) |
643b52b9 | 133 | { |
f8d5d0cc | 134 | root->xa_flags &= (__force gfp_t)~(1 << (tag + ROOT_TAG_SHIFT)); |
643b52b9 NP |
135 | } |
136 | ||
137 | static inline void root_tag_clear_all(struct radix_tree_root *root) | |
138 | { | |
f8d5d0cc | 139 | root->xa_flags &= (__force gfp_t)((1 << ROOT_TAG_SHIFT) - 1); |
643b52b9 NP |
140 | } |
141 | ||
35534c86 | 142 | static inline int root_tag_get(const struct radix_tree_root *root, unsigned tag) |
643b52b9 | 143 | { |
f8d5d0cc | 144 | return (__force int)root->xa_flags & (1 << (tag + ROOT_TAG_SHIFT)); |
643b52b9 NP |
145 | } |
146 | ||
35534c86 | 147 | static inline unsigned root_tags_get(const struct radix_tree_root *root) |
643b52b9 | 148 | { |
f8d5d0cc | 149 | return (__force unsigned)root->xa_flags >> ROOT_TAG_SHIFT; |
643b52b9 NP |
150 | } |
151 | ||
0a835c4f | 152 | static inline bool is_idr(const struct radix_tree_root *root) |
7b60e9ad | 153 | { |
f8d5d0cc | 154 | return !!(root->xa_flags & ROOT_IS_IDR); |
7b60e9ad MW |
155 | } |
156 | ||
643b52b9 NP |
157 | /* |
158 | * Returns 1 if any slot in the node has this tag set. | |
159 | * Otherwise returns 0. | |
160 | */ | |
35534c86 MW |
161 | static inline int any_tag_set(const struct radix_tree_node *node, |
162 | unsigned int tag) | |
643b52b9 | 163 | { |
2fcd9005 | 164 | unsigned idx; |
643b52b9 NP |
165 | for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) { |
166 | if (node->tags[tag][idx]) | |
167 | return 1; | |
168 | } | |
169 | return 0; | |
170 | } | |
78c1d784 | 171 | |
0a835c4f MW |
172 | static inline void all_tag_set(struct radix_tree_node *node, unsigned int tag) |
173 | { | |
174 | bitmap_fill(node->tags[tag], RADIX_TREE_MAP_SIZE); | |
175 | } | |
176 | ||
78c1d784 KK |
177 | /** |
178 | * radix_tree_find_next_bit - find the next set bit in a memory region | |
179 | * | |
180 | * @addr: The address to base the search on | |
181 | * @size: The bitmap size in bits | |
182 | * @offset: The bitnumber to start searching at | |
183 | * | |
184 | * Unrollable variant of find_next_bit() for constant size arrays. | |
185 | * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero. | |
186 | * Returns next bit offset, or size if nothing found. | |
187 | */ | |
188 | static __always_inline unsigned long | |
bc412fca MW |
189 | radix_tree_find_next_bit(struct radix_tree_node *node, unsigned int tag, |
190 | unsigned long offset) | |
78c1d784 | 191 | { |
bc412fca | 192 | const unsigned long *addr = node->tags[tag]; |
78c1d784 | 193 | |
bc412fca | 194 | if (offset < RADIX_TREE_MAP_SIZE) { |
78c1d784 KK |
195 | unsigned long tmp; |
196 | ||
197 | addr += offset / BITS_PER_LONG; | |
198 | tmp = *addr >> (offset % BITS_PER_LONG); | |
199 | if (tmp) | |
200 | return __ffs(tmp) + offset; | |
201 | offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1); | |
bc412fca | 202 | while (offset < RADIX_TREE_MAP_SIZE) { |
78c1d784 KK |
203 | tmp = *++addr; |
204 | if (tmp) | |
205 | return __ffs(tmp) + offset; | |
206 | offset += BITS_PER_LONG; | |
207 | } | |
208 | } | |
bc412fca | 209 | return RADIX_TREE_MAP_SIZE; |
78c1d784 KK |
210 | } |
211 | ||
268f42de MW |
212 | static unsigned int iter_offset(const struct radix_tree_iter *iter) |
213 | { | |
3a08cd52 | 214 | return iter->index & RADIX_TREE_MAP_MASK; |
268f42de MW |
215 | } |
216 | ||
218ed750 MW |
217 | /* |
218 | * The maximum index which can be stored in a radix tree | |
219 | */ | |
220 | static inline unsigned long shift_maxindex(unsigned int shift) | |
221 | { | |
222 | return (RADIX_TREE_MAP_SIZE << shift) - 1; | |
223 | } | |
224 | ||
35534c86 | 225 | static inline unsigned long node_maxindex(const struct radix_tree_node *node) |
218ed750 MW |
226 | { |
227 | return shift_maxindex(node->shift); | |
228 | } | |
229 | ||
0a835c4f MW |
230 | static unsigned long next_index(unsigned long index, |
231 | const struct radix_tree_node *node, | |
232 | unsigned long offset) | |
233 | { | |
234 | return (index & ~node_maxindex(node)) + (offset << node->shift); | |
235 | } | |
236 | ||
1da177e4 LT |
237 | /* |
238 | * This assumes that the caller has performed appropriate preallocation, and | |
239 | * that the caller has pinned this thread of control to the current CPU. | |
240 | */ | |
241 | static struct radix_tree_node * | |
0a835c4f | 242 | radix_tree_node_alloc(gfp_t gfp_mask, struct radix_tree_node *parent, |
d58275bc | 243 | struct radix_tree_root *root, |
e8de4340 | 244 | unsigned int shift, unsigned int offset, |
01959dfe | 245 | unsigned int count, unsigned int nr_values) |
1da177e4 | 246 | { |
e2848a0e | 247 | struct radix_tree_node *ret = NULL; |
1da177e4 | 248 | |
5e4c0d97 | 249 | /* |
2fcd9005 MW |
250 | * Preload code isn't irq safe and it doesn't make sense to use |
251 | * preloading during an interrupt anyway as all the allocations have | |
252 | * to be atomic. So just do normal allocation when in interrupt. | |
5e4c0d97 | 253 | */ |
d0164adc | 254 | if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) { |
1da177e4 LT |
255 | struct radix_tree_preload *rtp; |
256 | ||
58e698af VD |
257 | /* |
258 | * Even if the caller has preloaded, try to allocate from the | |
05eb6e72 VD |
259 | * cache first for the new node to get accounted to the memory |
260 | * cgroup. | |
58e698af VD |
261 | */ |
262 | ret = kmem_cache_alloc(radix_tree_node_cachep, | |
05eb6e72 | 263 | gfp_mask | __GFP_NOWARN); |
58e698af VD |
264 | if (ret) |
265 | goto out; | |
266 | ||
e2848a0e NP |
267 | /* |
268 | * Provided the caller has preloaded here, we will always | |
269 | * succeed in getting a node here (and never reach | |
270 | * kmem_cache_alloc) | |
271 | */ | |
7c8e0181 | 272 | rtp = this_cpu_ptr(&radix_tree_preloads); |
1da177e4 | 273 | if (rtp->nr) { |
9d2a8da0 | 274 | ret = rtp->nodes; |
1293d5c5 | 275 | rtp->nodes = ret->parent; |
1da177e4 LT |
276 | rtp->nr--; |
277 | } | |
ce80b067 CM |
278 | /* |
279 | * Update the allocation stack trace as this is more useful | |
280 | * for debugging. | |
281 | */ | |
282 | kmemleak_update_trace(ret); | |
58e698af | 283 | goto out; |
1da177e4 | 284 | } |
05eb6e72 | 285 | ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
58e698af | 286 | out: |
b194d16c | 287 | BUG_ON(radix_tree_is_internal_node(ret)); |
e8de4340 | 288 | if (ret) { |
e8de4340 MW |
289 | ret->shift = shift; |
290 | ret->offset = offset; | |
291 | ret->count = count; | |
01959dfe | 292 | ret->nr_values = nr_values; |
d58275bc | 293 | ret->parent = parent; |
01959dfe | 294 | ret->array = root; |
e8de4340 | 295 | } |
1da177e4 LT |
296 | return ret; |
297 | } | |
298 | ||
58d6ea30 | 299 | void radix_tree_node_rcu_free(struct rcu_head *head) |
7cf9c2c7 NP |
300 | { |
301 | struct radix_tree_node *node = | |
302 | container_of(head, struct radix_tree_node, rcu_head); | |
643b52b9 NP |
303 | |
304 | /* | |
175542f5 MW |
305 | * Must only free zeroed nodes into the slab. We can be left with |
306 | * non-NULL entries by radix_tree_free_nodes, so clear the entries | |
307 | * and tags here. | |
643b52b9 | 308 | */ |
175542f5 MW |
309 | memset(node->slots, 0, sizeof(node->slots)); |
310 | memset(node->tags, 0, sizeof(node->tags)); | |
91d9c05a | 311 | INIT_LIST_HEAD(&node->private_list); |
643b52b9 | 312 | |
7cf9c2c7 NP |
313 | kmem_cache_free(radix_tree_node_cachep, node); |
314 | } | |
315 | ||
1da177e4 LT |
316 | static inline void |
317 | radix_tree_node_free(struct radix_tree_node *node) | |
318 | { | |
7cf9c2c7 | 319 | call_rcu(&node->rcu_head, radix_tree_node_rcu_free); |
1da177e4 LT |
320 | } |
321 | ||
322 | /* | |
323 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
324 | * ensure that the addition of a single element in the tree cannot fail. On | |
325 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
326 | * with preemption not disabled. | |
b34df792 DH |
327 | * |
328 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 329 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
1da177e4 | 330 | */ |
bc9ae224 | 331 | static __must_check int __radix_tree_preload(gfp_t gfp_mask, unsigned nr) |
1da177e4 LT |
332 | { |
333 | struct radix_tree_preload *rtp; | |
334 | struct radix_tree_node *node; | |
335 | int ret = -ENOMEM; | |
336 | ||
05eb6e72 VD |
337 | /* |
338 | * Nodes preloaded by one cgroup can be be used by another cgroup, so | |
339 | * they should never be accounted to any particular memory cgroup. | |
340 | */ | |
341 | gfp_mask &= ~__GFP_ACCOUNT; | |
342 | ||
1da177e4 | 343 | preempt_disable(); |
7c8e0181 | 344 | rtp = this_cpu_ptr(&radix_tree_preloads); |
c78c66d1 | 345 | while (rtp->nr < nr) { |
1da177e4 | 346 | preempt_enable(); |
488514d1 | 347 | node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
1da177e4 LT |
348 | if (node == NULL) |
349 | goto out; | |
350 | preempt_disable(); | |
7c8e0181 | 351 | rtp = this_cpu_ptr(&radix_tree_preloads); |
c78c66d1 | 352 | if (rtp->nr < nr) { |
1293d5c5 | 353 | node->parent = rtp->nodes; |
9d2a8da0 KS |
354 | rtp->nodes = node; |
355 | rtp->nr++; | |
356 | } else { | |
1da177e4 | 357 | kmem_cache_free(radix_tree_node_cachep, node); |
9d2a8da0 | 358 | } |
1da177e4 LT |
359 | } |
360 | ret = 0; | |
361 | out: | |
362 | return ret; | |
363 | } | |
5e4c0d97 JK |
364 | |
365 | /* | |
366 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
367 | * ensure that the addition of a single element in the tree cannot fail. On | |
368 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
369 | * with preemption not disabled. | |
370 | * | |
371 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 372 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
5e4c0d97 JK |
373 | */ |
374 | int radix_tree_preload(gfp_t gfp_mask) | |
375 | { | |
376 | /* Warn on non-sensical use... */ | |
d0164adc | 377 | WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask)); |
c78c66d1 | 378 | return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE); |
5e4c0d97 | 379 | } |
d7f0923d | 380 | EXPORT_SYMBOL(radix_tree_preload); |
1da177e4 | 381 | |
5e4c0d97 JK |
382 | /* |
383 | * The same as above function, except we don't guarantee preloading happens. | |
384 | * We do it, if we decide it helps. On success, return zero with preemption | |
385 | * disabled. On error, return -ENOMEM with preemption not disabled. | |
386 | */ | |
387 | int radix_tree_maybe_preload(gfp_t gfp_mask) | |
388 | { | |
d0164adc | 389 | if (gfpflags_allow_blocking(gfp_mask)) |
c78c66d1 | 390 | return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE); |
5e4c0d97 JK |
391 | /* Preloading doesn't help anything with this gfp mask, skip it */ |
392 | preempt_disable(); | |
393 | return 0; | |
394 | } | |
395 | EXPORT_SYMBOL(radix_tree_maybe_preload); | |
396 | ||
35534c86 | 397 | static unsigned radix_tree_load_root(const struct radix_tree_root *root, |
1456a439 MW |
398 | struct radix_tree_node **nodep, unsigned long *maxindex) |
399 | { | |
f8d5d0cc | 400 | struct radix_tree_node *node = rcu_dereference_raw(root->xa_head); |
1456a439 MW |
401 | |
402 | *nodep = node; | |
403 | ||
b194d16c | 404 | if (likely(radix_tree_is_internal_node(node))) { |
4dd6c098 | 405 | node = entry_to_node(node); |
1456a439 | 406 | *maxindex = node_maxindex(node); |
c12e51b0 | 407 | return node->shift + RADIX_TREE_MAP_SHIFT; |
1456a439 MW |
408 | } |
409 | ||
410 | *maxindex = 0; | |
411 | return 0; | |
412 | } | |
413 | ||
1da177e4 LT |
414 | /* |
415 | * Extend a radix tree so it can store key @index. | |
416 | */ | |
0a835c4f | 417 | static int radix_tree_extend(struct radix_tree_root *root, gfp_t gfp, |
d0891265 | 418 | unsigned long index, unsigned int shift) |
1da177e4 | 419 | { |
d7b62727 | 420 | void *entry; |
d0891265 | 421 | unsigned int maxshift; |
1da177e4 LT |
422 | int tag; |
423 | ||
d0891265 MW |
424 | /* Figure out what the shift should be. */ |
425 | maxshift = shift; | |
426 | while (index > shift_maxindex(maxshift)) | |
427 | maxshift += RADIX_TREE_MAP_SHIFT; | |
1da177e4 | 428 | |
f8d5d0cc | 429 | entry = rcu_dereference_raw(root->xa_head); |
d7b62727 | 430 | if (!entry && (!is_idr(root) || root_tag_get(root, IDR_FREE))) |
1da177e4 | 431 | goto out; |
1da177e4 | 432 | |
1da177e4 | 433 | do { |
0a835c4f | 434 | struct radix_tree_node *node = radix_tree_node_alloc(gfp, NULL, |
d58275bc | 435 | root, shift, 0, 1, 0); |
2fcd9005 | 436 | if (!node) |
1da177e4 LT |
437 | return -ENOMEM; |
438 | ||
0a835c4f MW |
439 | if (is_idr(root)) { |
440 | all_tag_set(node, IDR_FREE); | |
441 | if (!root_tag_get(root, IDR_FREE)) { | |
442 | tag_clear(node, IDR_FREE, 0); | |
443 | root_tag_set(root, IDR_FREE); | |
444 | } | |
445 | } else { | |
446 | /* Propagate the aggregated tag info to the new child */ | |
447 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { | |
448 | if (root_tag_get(root, tag)) | |
449 | tag_set(node, tag, 0); | |
450 | } | |
1da177e4 LT |
451 | } |
452 | ||
d0891265 | 453 | BUG_ON(shift > BITS_PER_LONG); |
d7b62727 MW |
454 | if (radix_tree_is_internal_node(entry)) { |
455 | entry_to_node(entry)->parent = node; | |
3159f943 | 456 | } else if (xa_is_value(entry)) { |
01959dfe MW |
457 | /* Moving a value entry root->xa_head to a node */ |
458 | node->nr_values = 1; | |
f7942430 | 459 | } |
d7b62727 MW |
460 | /* |
461 | * entry was already in the radix tree, so we do not need | |
462 | * rcu_assign_pointer here | |
463 | */ | |
464 | node->slots[0] = (void __rcu *)entry; | |
465 | entry = node_to_entry(node); | |
f8d5d0cc | 466 | rcu_assign_pointer(root->xa_head, entry); |
d0891265 | 467 | shift += RADIX_TREE_MAP_SHIFT; |
d0891265 | 468 | } while (shift <= maxshift); |
1da177e4 | 469 | out: |
d0891265 | 470 | return maxshift + RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
471 | } |
472 | ||
f4b109c6 JW |
473 | /** |
474 | * radix_tree_shrink - shrink radix tree to minimum height | |
475 | * @root radix tree root | |
476 | */ | |
1cf56f9d | 477 | static inline bool radix_tree_shrink(struct radix_tree_root *root) |
f4b109c6 | 478 | { |
0ac398ef MW |
479 | bool shrunk = false; |
480 | ||
f4b109c6 | 481 | for (;;) { |
f8d5d0cc | 482 | struct radix_tree_node *node = rcu_dereference_raw(root->xa_head); |
f4b109c6 JW |
483 | struct radix_tree_node *child; |
484 | ||
485 | if (!radix_tree_is_internal_node(node)) | |
486 | break; | |
487 | node = entry_to_node(node); | |
488 | ||
489 | /* | |
490 | * The candidate node has more than one child, or its child | |
3a08cd52 | 491 | * is not at the leftmost slot, we cannot shrink. |
f4b109c6 JW |
492 | */ |
493 | if (node->count != 1) | |
494 | break; | |
12320d0f | 495 | child = rcu_dereference_raw(node->slots[0]); |
f4b109c6 JW |
496 | if (!child) |
497 | break; | |
f4b109c6 | 498 | |
66ee620f MW |
499 | /* |
500 | * For an IDR, we must not shrink entry 0 into the root in | |
501 | * case somebody calls idr_replace() with a pointer that | |
502 | * appears to be an internal entry | |
503 | */ | |
504 | if (!node->shift && is_idr(root)) | |
505 | break; | |
506 | ||
f4b109c6 JW |
507 | if (radix_tree_is_internal_node(child)) |
508 | entry_to_node(child)->parent = NULL; | |
509 | ||
510 | /* | |
511 | * We don't need rcu_assign_pointer(), since we are simply | |
512 | * moving the node from one part of the tree to another: if it | |
513 | * was safe to dereference the old pointer to it | |
514 | * (node->slots[0]), it will be safe to dereference the new | |
f8d5d0cc | 515 | * one (root->xa_head) as far as dependent read barriers go. |
f4b109c6 | 516 | */ |
f8d5d0cc | 517 | root->xa_head = (void __rcu *)child; |
0a835c4f MW |
518 | if (is_idr(root) && !tag_get(node, IDR_FREE, 0)) |
519 | root_tag_clear(root, IDR_FREE); | |
f4b109c6 JW |
520 | |
521 | /* | |
522 | * We have a dilemma here. The node's slot[0] must not be | |
523 | * NULLed in case there are concurrent lookups expecting to | |
524 | * find the item. However if this was a bottom-level node, | |
525 | * then it may be subject to the slot pointer being visible | |
526 | * to callers dereferencing it. If item corresponding to | |
527 | * slot[0] is subsequently deleted, these callers would expect | |
528 | * their slot to become empty sooner or later. | |
529 | * | |
530 | * For example, lockless pagecache will look up a slot, deref | |
531 | * the page pointer, and if the page has 0 refcount it means it | |
532 | * was concurrently deleted from pagecache so try the deref | |
533 | * again. Fortunately there is already a requirement for logic | |
534 | * to retry the entire slot lookup -- the indirect pointer | |
535 | * problem (replacing direct root node with an indirect pointer | |
536 | * also results in a stale slot). So tag the slot as indirect | |
537 | * to force callers to retry. | |
538 | */ | |
4d693d08 JW |
539 | node->count = 0; |
540 | if (!radix_tree_is_internal_node(child)) { | |
d7b62727 | 541 | node->slots[0] = (void __rcu *)RADIX_TREE_RETRY; |
4d693d08 | 542 | } |
f4b109c6 | 543 | |
ea07b862 | 544 | WARN_ON_ONCE(!list_empty(&node->private_list)); |
f4b109c6 | 545 | radix_tree_node_free(node); |
0ac398ef | 546 | shrunk = true; |
f4b109c6 | 547 | } |
0ac398ef MW |
548 | |
549 | return shrunk; | |
f4b109c6 JW |
550 | } |
551 | ||
0ac398ef | 552 | static bool delete_node(struct radix_tree_root *root, |
1cf56f9d | 553 | struct radix_tree_node *node) |
f4b109c6 | 554 | { |
0ac398ef MW |
555 | bool deleted = false; |
556 | ||
f4b109c6 JW |
557 | do { |
558 | struct radix_tree_node *parent; | |
559 | ||
560 | if (node->count) { | |
12320d0f | 561 | if (node_to_entry(node) == |
f8d5d0cc | 562 | rcu_dereference_raw(root->xa_head)) |
1cf56f9d | 563 | deleted |= radix_tree_shrink(root); |
0ac398ef | 564 | return deleted; |
f4b109c6 JW |
565 | } |
566 | ||
567 | parent = node->parent; | |
568 | if (parent) { | |
569 | parent->slots[node->offset] = NULL; | |
570 | parent->count--; | |
571 | } else { | |
0a835c4f MW |
572 | /* |
573 | * Shouldn't the tags already have all been cleared | |
574 | * by the caller? | |
575 | */ | |
576 | if (!is_idr(root)) | |
577 | root_tag_clear_all(root); | |
f8d5d0cc | 578 | root->xa_head = NULL; |
f4b109c6 JW |
579 | } |
580 | ||
ea07b862 | 581 | WARN_ON_ONCE(!list_empty(&node->private_list)); |
f4b109c6 | 582 | radix_tree_node_free(node); |
0ac398ef | 583 | deleted = true; |
f4b109c6 JW |
584 | |
585 | node = parent; | |
586 | } while (node); | |
0ac398ef MW |
587 | |
588 | return deleted; | |
f4b109c6 JW |
589 | } |
590 | ||
1da177e4 | 591 | /** |
139e5616 | 592 | * __radix_tree_create - create a slot in a radix tree |
1da177e4 LT |
593 | * @root: radix tree root |
594 | * @index: index key | |
139e5616 JW |
595 | * @nodep: returns node |
596 | * @slotp: returns slot | |
1da177e4 | 597 | * |
139e5616 JW |
598 | * Create, if necessary, and return the node and slot for an item |
599 | * at position @index in the radix tree @root. | |
600 | * | |
601 | * Until there is more than one item in the tree, no nodes are | |
f8d5d0cc | 602 | * allocated and @root->xa_head is used as a direct slot instead of |
139e5616 JW |
603 | * pointing to a node, in which case *@nodep will be NULL. |
604 | * | |
605 | * Returns -ENOMEM, or 0 for success. | |
1da177e4 | 606 | */ |
74d60958 | 607 | static int __radix_tree_create(struct radix_tree_root *root, |
3a08cd52 MW |
608 | unsigned long index, struct radix_tree_node **nodep, |
609 | void __rcu ***slotp) | |
1da177e4 | 610 | { |
89148aa4 | 611 | struct radix_tree_node *node = NULL, *child; |
f8d5d0cc | 612 | void __rcu **slot = (void __rcu **)&root->xa_head; |
49ea6ebc | 613 | unsigned long maxindex; |
89148aa4 | 614 | unsigned int shift, offset = 0; |
3a08cd52 | 615 | unsigned long max = index; |
0a835c4f | 616 | gfp_t gfp = root_gfp_mask(root); |
49ea6ebc | 617 | |
89148aa4 | 618 | shift = radix_tree_load_root(root, &child, &maxindex); |
1da177e4 LT |
619 | |
620 | /* Make sure the tree is high enough. */ | |
49ea6ebc | 621 | if (max > maxindex) { |
0a835c4f | 622 | int error = radix_tree_extend(root, gfp, max, shift); |
49ea6ebc | 623 | if (error < 0) |
1da177e4 | 624 | return error; |
49ea6ebc | 625 | shift = error; |
f8d5d0cc | 626 | child = rcu_dereference_raw(root->xa_head); |
1da177e4 LT |
627 | } |
628 | ||
3a08cd52 | 629 | while (shift > 0) { |
c12e51b0 | 630 | shift -= RADIX_TREE_MAP_SHIFT; |
89148aa4 | 631 | if (child == NULL) { |
1da177e4 | 632 | /* Have to add a child node. */ |
d58275bc | 633 | child = radix_tree_node_alloc(gfp, node, root, shift, |
e8de4340 | 634 | offset, 0, 0); |
89148aa4 | 635 | if (!child) |
1da177e4 | 636 | return -ENOMEM; |
89148aa4 MW |
637 | rcu_assign_pointer(*slot, node_to_entry(child)); |
638 | if (node) | |
1da177e4 | 639 | node->count++; |
89148aa4 | 640 | } else if (!radix_tree_is_internal_node(child)) |
e6145236 | 641 | break; |
1da177e4 LT |
642 | |
643 | /* Go a level down */ | |
89148aa4 | 644 | node = entry_to_node(child); |
9e85d811 | 645 | offset = radix_tree_descend(node, &child, index); |
89148aa4 | 646 | slot = &node->slots[offset]; |
e6145236 MW |
647 | } |
648 | ||
175542f5 MW |
649 | if (nodep) |
650 | *nodep = node; | |
651 | if (slotp) | |
652 | *slotp = slot; | |
653 | return 0; | |
654 | } | |
655 | ||
175542f5 MW |
656 | /* |
657 | * Free any nodes below this node. The tree is presumed to not need | |
658 | * shrinking, and any user data in the tree is presumed to not need a | |
659 | * destructor called on it. If we need to add a destructor, we can | |
660 | * add that functionality later. Note that we may not clear tags or | |
661 | * slots from the tree as an RCU walker may still have a pointer into | |
662 | * this subtree. We could replace the entries with RADIX_TREE_RETRY, | |
663 | * but we'll still have to clear those in rcu_free. | |
664 | */ | |
665 | static void radix_tree_free_nodes(struct radix_tree_node *node) | |
666 | { | |
667 | unsigned offset = 0; | |
668 | struct radix_tree_node *child = entry_to_node(node); | |
669 | ||
670 | for (;;) { | |
12320d0f | 671 | void *entry = rcu_dereference_raw(child->slots[offset]); |
02c02bf1 | 672 | if (xa_is_node(entry) && child->shift) { |
175542f5 MW |
673 | child = entry_to_node(entry); |
674 | offset = 0; | |
675 | continue; | |
676 | } | |
677 | offset++; | |
678 | while (offset == RADIX_TREE_MAP_SIZE) { | |
679 | struct radix_tree_node *old = child; | |
680 | offset = child->offset + 1; | |
681 | child = child->parent; | |
dd040b6f | 682 | WARN_ON_ONCE(!list_empty(&old->private_list)); |
175542f5 MW |
683 | radix_tree_node_free(old); |
684 | if (old == entry_to_node(node)) | |
685 | return; | |
686 | } | |
687 | } | |
688 | } | |
689 | ||
d7b62727 | 690 | static inline int insert_entries(struct radix_tree_node *node, |
3a08cd52 | 691 | void __rcu **slot, void *item, bool replace) |
175542f5 MW |
692 | { |
693 | if (*slot) | |
694 | return -EEXIST; | |
695 | rcu_assign_pointer(*slot, item); | |
696 | if (node) { | |
697 | node->count++; | |
3159f943 | 698 | if (xa_is_value(item)) |
01959dfe | 699 | node->nr_values++; |
175542f5 MW |
700 | } |
701 | return 1; | |
702 | } | |
139e5616 JW |
703 | |
704 | /** | |
e6145236 | 705 | * __radix_tree_insert - insert into a radix tree |
139e5616 JW |
706 | * @root: radix tree root |
707 | * @index: index key | |
708 | * @item: item to insert | |
709 | * | |
710 | * Insert an item into the radix tree at position @index. | |
711 | */ | |
3a08cd52 MW |
712 | int radix_tree_insert(struct radix_tree_root *root, unsigned long index, |
713 | void *item) | |
139e5616 JW |
714 | { |
715 | struct radix_tree_node *node; | |
d7b62727 | 716 | void __rcu **slot; |
139e5616 JW |
717 | int error; |
718 | ||
b194d16c | 719 | BUG_ON(radix_tree_is_internal_node(item)); |
139e5616 | 720 | |
3a08cd52 | 721 | error = __radix_tree_create(root, index, &node, &slot); |
139e5616 JW |
722 | if (error) |
723 | return error; | |
175542f5 | 724 | |
3a08cd52 | 725 | error = insert_entries(node, slot, item, false); |
175542f5 MW |
726 | if (error < 0) |
727 | return error; | |
201b6264 | 728 | |
612d6c19 | 729 | if (node) { |
7b60e9ad | 730 | unsigned offset = get_slot_offset(node, slot); |
7b60e9ad MW |
731 | BUG_ON(tag_get(node, 0, offset)); |
732 | BUG_ON(tag_get(node, 1, offset)); | |
733 | BUG_ON(tag_get(node, 2, offset)); | |
612d6c19 | 734 | } else { |
7b60e9ad | 735 | BUG_ON(root_tags_get(root)); |
612d6c19 | 736 | } |
1da177e4 | 737 | |
1da177e4 LT |
738 | return 0; |
739 | } | |
3a08cd52 | 740 | EXPORT_SYMBOL(radix_tree_insert); |
1da177e4 | 741 | |
139e5616 JW |
742 | /** |
743 | * __radix_tree_lookup - lookup an item in a radix tree | |
744 | * @root: radix tree root | |
745 | * @index: index key | |
746 | * @nodep: returns node | |
747 | * @slotp: returns slot | |
748 | * | |
749 | * Lookup and return the item at position @index in the radix | |
750 | * tree @root. | |
751 | * | |
752 | * Until there is more than one item in the tree, no nodes are | |
f8d5d0cc | 753 | * allocated and @root->xa_head is used as a direct slot instead of |
139e5616 | 754 | * pointing to a node, in which case *@nodep will be NULL. |
7cf9c2c7 | 755 | */ |
35534c86 MW |
756 | void *__radix_tree_lookup(const struct radix_tree_root *root, |
757 | unsigned long index, struct radix_tree_node **nodep, | |
d7b62727 | 758 | void __rcu ***slotp) |
1da177e4 | 759 | { |
139e5616 | 760 | struct radix_tree_node *node, *parent; |
85829954 | 761 | unsigned long maxindex; |
d7b62727 | 762 | void __rcu **slot; |
612d6c19 | 763 | |
85829954 MW |
764 | restart: |
765 | parent = NULL; | |
f8d5d0cc | 766 | slot = (void __rcu **)&root->xa_head; |
9e85d811 | 767 | radix_tree_load_root(root, &node, &maxindex); |
85829954 | 768 | if (index > maxindex) |
1da177e4 LT |
769 | return NULL; |
770 | ||
b194d16c | 771 | while (radix_tree_is_internal_node(node)) { |
85829954 | 772 | unsigned offset; |
1da177e4 | 773 | |
4dd6c098 | 774 | parent = entry_to_node(node); |
9e85d811 | 775 | offset = radix_tree_descend(parent, &node, index); |
85829954 | 776 | slot = parent->slots + offset; |
eff3860b MW |
777 | if (node == RADIX_TREE_RETRY) |
778 | goto restart; | |
66ee620f MW |
779 | if (parent->shift == 0) |
780 | break; | |
85829954 | 781 | } |
1da177e4 | 782 | |
139e5616 JW |
783 | if (nodep) |
784 | *nodep = parent; | |
785 | if (slotp) | |
786 | *slotp = slot; | |
787 | return node; | |
b72b71c6 HS |
788 | } |
789 | ||
790 | /** | |
791 | * radix_tree_lookup_slot - lookup a slot in a radix tree | |
792 | * @root: radix tree root | |
793 | * @index: index key | |
794 | * | |
795 | * Returns: the slot corresponding to the position @index in the | |
796 | * radix tree @root. This is useful for update-if-exists operations. | |
797 | * | |
798 | * This function can be called under rcu_read_lock iff the slot is not | |
799 | * modified by radix_tree_replace_slot, otherwise it must be called | |
800 | * exclusive from other writers. Any dereference of the slot must be done | |
801 | * using radix_tree_deref_slot. | |
802 | */ | |
d7b62727 | 803 | void __rcu **radix_tree_lookup_slot(const struct radix_tree_root *root, |
35534c86 | 804 | unsigned long index) |
b72b71c6 | 805 | { |
d7b62727 | 806 | void __rcu **slot; |
139e5616 JW |
807 | |
808 | if (!__radix_tree_lookup(root, index, NULL, &slot)) | |
809 | return NULL; | |
810 | return slot; | |
a4331366 | 811 | } |
a4331366 HR |
812 | EXPORT_SYMBOL(radix_tree_lookup_slot); |
813 | ||
814 | /** | |
815 | * radix_tree_lookup - perform lookup operation on a radix tree | |
816 | * @root: radix tree root | |
817 | * @index: index key | |
818 | * | |
819 | * Lookup the item at the position @index in the radix tree @root. | |
7cf9c2c7 NP |
820 | * |
821 | * This function can be called under rcu_read_lock, however the caller | |
822 | * must manage lifetimes of leaf nodes (eg. RCU may also be used to free | |
823 | * them safely). No RCU barriers are required to access or modify the | |
824 | * returned item, however. | |
a4331366 | 825 | */ |
35534c86 | 826 | void *radix_tree_lookup(const struct radix_tree_root *root, unsigned long index) |
a4331366 | 827 | { |
139e5616 | 828 | return __radix_tree_lookup(root, index, NULL, NULL); |
1da177e4 LT |
829 | } |
830 | EXPORT_SYMBOL(radix_tree_lookup); | |
831 | ||
d7b62727 | 832 | static void replace_slot(void __rcu **slot, void *item, |
01959dfe | 833 | struct radix_tree_node *node, int count, int values) |
f7942430 | 834 | { |
01959dfe | 835 | if (node && (count || values)) { |
f4b109c6 | 836 | node->count += count; |
01959dfe | 837 | node->nr_values += values; |
f4b109c6 | 838 | } |
f7942430 JW |
839 | |
840 | rcu_assign_pointer(*slot, item); | |
841 | } | |
842 | ||
0a835c4f MW |
843 | static bool node_tag_get(const struct radix_tree_root *root, |
844 | const struct radix_tree_node *node, | |
845 | unsigned int tag, unsigned int offset) | |
a90eb3a2 | 846 | { |
0a835c4f MW |
847 | if (node) |
848 | return tag_get(node, tag, offset); | |
849 | return root_tag_get(root, tag); | |
850 | } | |
a90eb3a2 | 851 | |
0a835c4f MW |
852 | /* |
853 | * IDR users want to be able to store NULL in the tree, so if the slot isn't | |
854 | * free, don't adjust the count, even if it's transitioning between NULL and | |
855 | * non-NULL. For the IDA, we mark slots as being IDR_FREE while they still | |
856 | * have empty bits, but it only stores NULL in slots when they're being | |
857 | * deleted. | |
858 | */ | |
859 | static int calculate_count(struct radix_tree_root *root, | |
d7b62727 | 860 | struct radix_tree_node *node, void __rcu **slot, |
0a835c4f MW |
861 | void *item, void *old) |
862 | { | |
863 | if (is_idr(root)) { | |
864 | unsigned offset = get_slot_offset(node, slot); | |
865 | bool free = node_tag_get(root, node, IDR_FREE, offset); | |
866 | if (!free) | |
867 | return 0; | |
868 | if (!old) | |
869 | return 1; | |
a90eb3a2 | 870 | } |
0a835c4f | 871 | return !!item - !!old; |
a90eb3a2 MW |
872 | } |
873 | ||
6d75f366 JW |
874 | /** |
875 | * __radix_tree_replace - replace item in a slot | |
4d693d08 JW |
876 | * @root: radix tree root |
877 | * @node: pointer to tree node | |
878 | * @slot: pointer to slot in @node | |
879 | * @item: new item to store in the slot. | |
6d75f366 JW |
880 | * |
881 | * For use with __radix_tree_lookup(). Caller must hold tree write locked | |
882 | * across slot lookup and replacement. | |
883 | */ | |
884 | void __radix_tree_replace(struct radix_tree_root *root, | |
885 | struct radix_tree_node *node, | |
1cf56f9d | 886 | void __rcu **slot, void *item) |
6d75f366 | 887 | { |
0a835c4f | 888 | void *old = rcu_dereference_raw(*slot); |
01959dfe | 889 | int values = !!xa_is_value(item) - !!xa_is_value(old); |
0a835c4f MW |
890 | int count = calculate_count(root, node, slot, item, old); |
891 | ||
6d75f366 | 892 | /* |
01959dfe | 893 | * This function supports replacing value entries and |
f4b109c6 | 894 | * deleting entries, but that needs accounting against the |
f8d5d0cc | 895 | * node unless the slot is root->xa_head. |
6d75f366 | 896 | */ |
f8d5d0cc | 897 | WARN_ON_ONCE(!node && (slot != (void __rcu **)&root->xa_head) && |
01959dfe MW |
898 | (count || values)); |
899 | replace_slot(slot, item, node, count, values); | |
f4b109c6 | 900 | |
4d693d08 JW |
901 | if (!node) |
902 | return; | |
903 | ||
1cf56f9d | 904 | delete_node(root, node); |
6d75f366 JW |
905 | } |
906 | ||
907 | /** | |
908 | * radix_tree_replace_slot - replace item in a slot | |
909 | * @root: radix tree root | |
910 | * @slot: pointer to slot | |
911 | * @item: new item to store in the slot. | |
912 | * | |
7b8d046f | 913 | * For use with radix_tree_lookup_slot() and |
6d75f366 JW |
914 | * radix_tree_gang_lookup_tag_slot(). Caller must hold tree write locked |
915 | * across slot lookup and replacement. | |
916 | * | |
917 | * NOTE: This cannot be used to switch between non-entries (empty slots), | |
01959dfe | 918 | * regular entries, and value entries, as that requires accounting |
f4b109c6 | 919 | * inside the radix tree node. When switching from one type of entry or |
e157b555 MW |
920 | * deleting, use __radix_tree_lookup() and __radix_tree_replace() or |
921 | * radix_tree_iter_replace(). | |
6d75f366 JW |
922 | */ |
923 | void radix_tree_replace_slot(struct radix_tree_root *root, | |
d7b62727 | 924 | void __rcu **slot, void *item) |
6d75f366 | 925 | { |
1cf56f9d | 926 | __radix_tree_replace(root, NULL, slot, item); |
6d75f366 | 927 | } |
10257d71 | 928 | EXPORT_SYMBOL(radix_tree_replace_slot); |
6d75f366 | 929 | |
e157b555 MW |
930 | /** |
931 | * radix_tree_iter_replace - replace item in a slot | |
932 | * @root: radix tree root | |
933 | * @slot: pointer to slot | |
934 | * @item: new item to store in the slot. | |
935 | * | |
2956c664 MW |
936 | * For use with radix_tree_for_each_slot(). |
937 | * Caller must hold tree write locked. | |
e157b555 MW |
938 | */ |
939 | void radix_tree_iter_replace(struct radix_tree_root *root, | |
d7b62727 MW |
940 | const struct radix_tree_iter *iter, |
941 | void __rcu **slot, void *item) | |
e157b555 | 942 | { |
1cf56f9d | 943 | __radix_tree_replace(root, iter->node, slot, item); |
e157b555 MW |
944 | } |
945 | ||
30b888ba MW |
946 | static void node_tag_set(struct radix_tree_root *root, |
947 | struct radix_tree_node *node, | |
948 | unsigned int tag, unsigned int offset) | |
949 | { | |
950 | while (node) { | |
951 | if (tag_get(node, tag, offset)) | |
952 | return; | |
953 | tag_set(node, tag, offset); | |
954 | offset = node->offset; | |
955 | node = node->parent; | |
956 | } | |
957 | ||
958 | if (!root_tag_get(root, tag)) | |
959 | root_tag_set(root, tag); | |
960 | } | |
961 | ||
1da177e4 LT |
962 | /** |
963 | * radix_tree_tag_set - set a tag on a radix tree node | |
964 | * @root: radix tree root | |
965 | * @index: index key | |
2fcd9005 | 966 | * @tag: tag index |
1da177e4 | 967 | * |
daff89f3 JC |
968 | * Set the search tag (which must be < RADIX_TREE_MAX_TAGS) |
969 | * corresponding to @index in the radix tree. From | |
1da177e4 LT |
970 | * the root all the way down to the leaf node. |
971 | * | |
2fcd9005 | 972 | * Returns the address of the tagged item. Setting a tag on a not-present |
1da177e4 LT |
973 | * item is a bug. |
974 | */ | |
975 | void *radix_tree_tag_set(struct radix_tree_root *root, | |
daff89f3 | 976 | unsigned long index, unsigned int tag) |
1da177e4 | 977 | { |
fb969909 RZ |
978 | struct radix_tree_node *node, *parent; |
979 | unsigned long maxindex; | |
1da177e4 | 980 | |
9e85d811 | 981 | radix_tree_load_root(root, &node, &maxindex); |
fb969909 | 982 | BUG_ON(index > maxindex); |
1da177e4 | 983 | |
b194d16c | 984 | while (radix_tree_is_internal_node(node)) { |
fb969909 | 985 | unsigned offset; |
1da177e4 | 986 | |
4dd6c098 | 987 | parent = entry_to_node(node); |
9e85d811 | 988 | offset = radix_tree_descend(parent, &node, index); |
fb969909 RZ |
989 | BUG_ON(!node); |
990 | ||
991 | if (!tag_get(parent, tag, offset)) | |
992 | tag_set(parent, tag, offset); | |
1da177e4 LT |
993 | } |
994 | ||
612d6c19 | 995 | /* set the root's tag bit */ |
fb969909 | 996 | if (!root_tag_get(root, tag)) |
612d6c19 NP |
997 | root_tag_set(root, tag); |
998 | ||
fb969909 | 999 | return node; |
1da177e4 LT |
1000 | } |
1001 | EXPORT_SYMBOL(radix_tree_tag_set); | |
1002 | ||
d604c324 MW |
1003 | static void node_tag_clear(struct radix_tree_root *root, |
1004 | struct radix_tree_node *node, | |
1005 | unsigned int tag, unsigned int offset) | |
1006 | { | |
1007 | while (node) { | |
1008 | if (!tag_get(node, tag, offset)) | |
1009 | return; | |
1010 | tag_clear(node, tag, offset); | |
1011 | if (any_tag_set(node, tag)) | |
1012 | return; | |
1013 | ||
1014 | offset = node->offset; | |
1015 | node = node->parent; | |
1016 | } | |
1017 | ||
1018 | /* clear the root's tag bit */ | |
1019 | if (root_tag_get(root, tag)) | |
1020 | root_tag_clear(root, tag); | |
1021 | } | |
1022 | ||
1da177e4 LT |
1023 | /** |
1024 | * radix_tree_tag_clear - clear a tag on a radix tree node | |
1025 | * @root: radix tree root | |
1026 | * @index: index key | |
2fcd9005 | 1027 | * @tag: tag index |
1da177e4 | 1028 | * |
daff89f3 | 1029 | * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS) |
2fcd9005 MW |
1030 | * corresponding to @index in the radix tree. If this causes |
1031 | * the leaf node to have no tags set then clear the tag in the | |
1da177e4 LT |
1032 | * next-to-leaf node, etc. |
1033 | * | |
1034 | * Returns the address of the tagged item on success, else NULL. ie: | |
1035 | * has the same return value and semantics as radix_tree_lookup(). | |
1036 | */ | |
1037 | void *radix_tree_tag_clear(struct radix_tree_root *root, | |
daff89f3 | 1038 | unsigned long index, unsigned int tag) |
1da177e4 | 1039 | { |
00f47b58 RZ |
1040 | struct radix_tree_node *node, *parent; |
1041 | unsigned long maxindex; | |
e2bdb933 | 1042 | int uninitialized_var(offset); |
1da177e4 | 1043 | |
9e85d811 | 1044 | radix_tree_load_root(root, &node, &maxindex); |
00f47b58 RZ |
1045 | if (index > maxindex) |
1046 | return NULL; | |
1da177e4 | 1047 | |
00f47b58 | 1048 | parent = NULL; |
1da177e4 | 1049 | |
b194d16c | 1050 | while (radix_tree_is_internal_node(node)) { |
4dd6c098 | 1051 | parent = entry_to_node(node); |
9e85d811 | 1052 | offset = radix_tree_descend(parent, &node, index); |
1da177e4 LT |
1053 | } |
1054 | ||
d604c324 MW |
1055 | if (node) |
1056 | node_tag_clear(root, parent, tag, offset); | |
1da177e4 | 1057 | |
00f47b58 | 1058 | return node; |
1da177e4 LT |
1059 | } |
1060 | EXPORT_SYMBOL(radix_tree_tag_clear); | |
1061 | ||
30b888ba MW |
1062 | /** |
1063 | * radix_tree_iter_tag_clear - clear a tag on the current iterator entry | |
1064 | * @root: radix tree root | |
1065 | * @iter: iterator state | |
1066 | * @tag: tag to clear | |
1067 | */ | |
1068 | void radix_tree_iter_tag_clear(struct radix_tree_root *root, | |
1069 | const struct radix_tree_iter *iter, unsigned int tag) | |
1070 | { | |
1071 | node_tag_clear(root, iter->node, tag, iter_offset(iter)); | |
1072 | } | |
1073 | ||
1da177e4 | 1074 | /** |
32605a18 MT |
1075 | * radix_tree_tag_get - get a tag on a radix tree node |
1076 | * @root: radix tree root | |
1077 | * @index: index key | |
2fcd9005 | 1078 | * @tag: tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 | 1079 | * |
32605a18 | 1080 | * Return values: |
1da177e4 | 1081 | * |
612d6c19 NP |
1082 | * 0: tag not present or not set |
1083 | * 1: tag set | |
ce82653d DH |
1084 | * |
1085 | * Note that the return value of this function may not be relied on, even if | |
1086 | * the RCU lock is held, unless tag modification and node deletion are excluded | |
1087 | * from concurrency. | |
1da177e4 | 1088 | */ |
35534c86 | 1089 | int radix_tree_tag_get(const struct radix_tree_root *root, |
daff89f3 | 1090 | unsigned long index, unsigned int tag) |
1da177e4 | 1091 | { |
4589ba6d RZ |
1092 | struct radix_tree_node *node, *parent; |
1093 | unsigned long maxindex; | |
1da177e4 | 1094 | |
612d6c19 NP |
1095 | if (!root_tag_get(root, tag)) |
1096 | return 0; | |
1097 | ||
9e85d811 | 1098 | radix_tree_load_root(root, &node, &maxindex); |
4589ba6d RZ |
1099 | if (index > maxindex) |
1100 | return 0; | |
7cf9c2c7 | 1101 | |
b194d16c | 1102 | while (radix_tree_is_internal_node(node)) { |
9e85d811 | 1103 | unsigned offset; |
1da177e4 | 1104 | |
4dd6c098 | 1105 | parent = entry_to_node(node); |
9e85d811 | 1106 | offset = radix_tree_descend(parent, &node, index); |
1da177e4 | 1107 | |
4589ba6d | 1108 | if (!tag_get(parent, tag, offset)) |
3fa36acb | 1109 | return 0; |
4589ba6d RZ |
1110 | if (node == RADIX_TREE_RETRY) |
1111 | break; | |
1da177e4 | 1112 | } |
4589ba6d RZ |
1113 | |
1114 | return 1; | |
1da177e4 LT |
1115 | } |
1116 | EXPORT_SYMBOL(radix_tree_tag_get); | |
1da177e4 | 1117 | |
148deab2 MW |
1118 | /* Construct iter->tags bit-mask from node->tags[tag] array */ |
1119 | static void set_iter_tags(struct radix_tree_iter *iter, | |
1120 | struct radix_tree_node *node, unsigned offset, | |
1121 | unsigned tag) | |
1122 | { | |
1123 | unsigned tag_long = offset / BITS_PER_LONG; | |
1124 | unsigned tag_bit = offset % BITS_PER_LONG; | |
1125 | ||
0a835c4f MW |
1126 | if (!node) { |
1127 | iter->tags = 1; | |
1128 | return; | |
1129 | } | |
1130 | ||
148deab2 MW |
1131 | iter->tags = node->tags[tag][tag_long] >> tag_bit; |
1132 | ||
1133 | /* This never happens if RADIX_TREE_TAG_LONGS == 1 */ | |
1134 | if (tag_long < RADIX_TREE_TAG_LONGS - 1) { | |
1135 | /* Pick tags from next element */ | |
1136 | if (tag_bit) | |
1137 | iter->tags |= node->tags[tag][tag_long + 1] << | |
1138 | (BITS_PER_LONG - tag_bit); | |
1139 | /* Clip chunk size, here only BITS_PER_LONG tags */ | |
1140 | iter->next_index = __radix_tree_iter_add(iter, BITS_PER_LONG); | |
1141 | } | |
1142 | } | |
1143 | ||
d7b62727 MW |
1144 | void __rcu **radix_tree_iter_resume(void __rcu **slot, |
1145 | struct radix_tree_iter *iter) | |
148deab2 | 1146 | { |
148deab2 MW |
1147 | slot++; |
1148 | iter->index = __radix_tree_iter_add(iter, 1); | |
148deab2 MW |
1149 | iter->next_index = iter->index; |
1150 | iter->tags = 0; | |
1151 | return NULL; | |
1152 | } | |
1153 | EXPORT_SYMBOL(radix_tree_iter_resume); | |
1154 | ||
78c1d784 KK |
1155 | /** |
1156 | * radix_tree_next_chunk - find next chunk of slots for iteration | |
1157 | * | |
1158 | * @root: radix tree root | |
1159 | * @iter: iterator state | |
1160 | * @flags: RADIX_TREE_ITER_* flags and tag index | |
1161 | * Returns: pointer to chunk first slot, or NULL if iteration is over | |
1162 | */ | |
d7b62727 | 1163 | void __rcu **radix_tree_next_chunk(const struct radix_tree_root *root, |
78c1d784 KK |
1164 | struct radix_tree_iter *iter, unsigned flags) |
1165 | { | |
9e85d811 | 1166 | unsigned tag = flags & RADIX_TREE_ITER_TAG_MASK; |
8c1244de | 1167 | struct radix_tree_node *node, *child; |
21ef5339 | 1168 | unsigned long index, offset, maxindex; |
78c1d784 KK |
1169 | |
1170 | if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag)) | |
1171 | return NULL; | |
1172 | ||
1173 | /* | |
1174 | * Catch next_index overflow after ~0UL. iter->index never overflows | |
1175 | * during iterating; it can be zero only at the beginning. | |
1176 | * And we cannot overflow iter->next_index in a single step, | |
1177 | * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG. | |
fffaee36 KK |
1178 | * |
1179 | * This condition also used by radix_tree_next_slot() to stop | |
91b9677c | 1180 | * contiguous iterating, and forbid switching to the next chunk. |
78c1d784 KK |
1181 | */ |
1182 | index = iter->next_index; | |
1183 | if (!index && iter->index) | |
1184 | return NULL; | |
1185 | ||
21ef5339 | 1186 | restart: |
9e85d811 | 1187 | radix_tree_load_root(root, &child, &maxindex); |
21ef5339 RZ |
1188 | if (index > maxindex) |
1189 | return NULL; | |
8c1244de MW |
1190 | if (!child) |
1191 | return NULL; | |
21ef5339 | 1192 | |
8c1244de | 1193 | if (!radix_tree_is_internal_node(child)) { |
78c1d784 | 1194 | /* Single-slot tree */ |
21ef5339 RZ |
1195 | iter->index = index; |
1196 | iter->next_index = maxindex + 1; | |
78c1d784 | 1197 | iter->tags = 1; |
268f42de | 1198 | iter->node = NULL; |
f8d5d0cc | 1199 | return (void __rcu **)&root->xa_head; |
8c1244de | 1200 | } |
21ef5339 | 1201 | |
8c1244de MW |
1202 | do { |
1203 | node = entry_to_node(child); | |
9e85d811 | 1204 | offset = radix_tree_descend(node, &child, index); |
21ef5339 | 1205 | |
78c1d784 | 1206 | if ((flags & RADIX_TREE_ITER_TAGGED) ? |
8c1244de | 1207 | !tag_get(node, tag, offset) : !child) { |
78c1d784 KK |
1208 | /* Hole detected */ |
1209 | if (flags & RADIX_TREE_ITER_CONTIG) | |
1210 | return NULL; | |
1211 | ||
1212 | if (flags & RADIX_TREE_ITER_TAGGED) | |
bc412fca | 1213 | offset = radix_tree_find_next_bit(node, tag, |
78c1d784 KK |
1214 | offset + 1); |
1215 | else | |
1216 | while (++offset < RADIX_TREE_MAP_SIZE) { | |
12320d0f MW |
1217 | void *slot = rcu_dereference_raw( |
1218 | node->slots[offset]); | |
21ef5339 | 1219 | if (slot) |
78c1d784 KK |
1220 | break; |
1221 | } | |
8c1244de | 1222 | index &= ~node_maxindex(node); |
9e85d811 | 1223 | index += offset << node->shift; |
78c1d784 KK |
1224 | /* Overflow after ~0UL */ |
1225 | if (!index) | |
1226 | return NULL; | |
1227 | if (offset == RADIX_TREE_MAP_SIZE) | |
1228 | goto restart; | |
8c1244de | 1229 | child = rcu_dereference_raw(node->slots[offset]); |
78c1d784 KK |
1230 | } |
1231 | ||
e157b555 | 1232 | if (!child) |
78c1d784 | 1233 | goto restart; |
e157b555 MW |
1234 | if (child == RADIX_TREE_RETRY) |
1235 | break; | |
66ee620f | 1236 | } while (node->shift && radix_tree_is_internal_node(child)); |
78c1d784 KK |
1237 | |
1238 | /* Update the iterator state */ | |
3a08cd52 | 1239 | iter->index = (index &~ node_maxindex(node)) | offset; |
8c1244de | 1240 | iter->next_index = (index | node_maxindex(node)) + 1; |
268f42de | 1241 | iter->node = node; |
78c1d784 | 1242 | |
148deab2 MW |
1243 | if (flags & RADIX_TREE_ITER_TAGGED) |
1244 | set_iter_tags(iter, node, offset, tag); | |
78c1d784 KK |
1245 | |
1246 | return node->slots + offset; | |
1247 | } | |
1248 | EXPORT_SYMBOL(radix_tree_next_chunk); | |
1249 | ||
1da177e4 LT |
1250 | /** |
1251 | * radix_tree_gang_lookup - perform multiple lookup on a radix tree | |
1252 | * @root: radix tree root | |
1253 | * @results: where the results of the lookup are placed | |
1254 | * @first_index: start the lookup from this key | |
1255 | * @max_items: place up to this many items at *results | |
1256 | * | |
1257 | * Performs an index-ascending scan of the tree for present items. Places | |
1258 | * them at *@results and returns the number of items which were placed at | |
1259 | * *@results. | |
1260 | * | |
1261 | * The implementation is naive. | |
7cf9c2c7 NP |
1262 | * |
1263 | * Like radix_tree_lookup, radix_tree_gang_lookup may be called under | |
1264 | * rcu_read_lock. In this case, rather than the returned results being | |
2fcd9005 MW |
1265 | * an atomic snapshot of the tree at a single point in time, the |
1266 | * semantics of an RCU protected gang lookup are as though multiple | |
1267 | * radix_tree_lookups have been issued in individual locks, and results | |
1268 | * stored in 'results'. | |
1da177e4 LT |
1269 | */ |
1270 | unsigned int | |
35534c86 | 1271 | radix_tree_gang_lookup(const struct radix_tree_root *root, void **results, |
1da177e4 LT |
1272 | unsigned long first_index, unsigned int max_items) |
1273 | { | |
cebbd29e | 1274 | struct radix_tree_iter iter; |
d7b62727 | 1275 | void __rcu **slot; |
cebbd29e | 1276 | unsigned int ret = 0; |
7cf9c2c7 | 1277 | |
cebbd29e | 1278 | if (unlikely(!max_items)) |
7cf9c2c7 | 1279 | return 0; |
1da177e4 | 1280 | |
cebbd29e | 1281 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
46437f9a | 1282 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1283 | if (!results[ret]) |
1284 | continue; | |
b194d16c | 1285 | if (radix_tree_is_internal_node(results[ret])) { |
46437f9a MW |
1286 | slot = radix_tree_iter_retry(&iter); |
1287 | continue; | |
1288 | } | |
cebbd29e | 1289 | if (++ret == max_items) |
1da177e4 | 1290 | break; |
1da177e4 | 1291 | } |
7cf9c2c7 | 1292 | |
1da177e4 LT |
1293 | return ret; |
1294 | } | |
1295 | EXPORT_SYMBOL(radix_tree_gang_lookup); | |
1296 | ||
1da177e4 LT |
1297 | /** |
1298 | * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree | |
1299 | * based on a tag | |
1300 | * @root: radix tree root | |
1301 | * @results: where the results of the lookup are placed | |
1302 | * @first_index: start the lookup from this key | |
1303 | * @max_items: place up to this many items at *results | |
daff89f3 | 1304 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 LT |
1305 | * |
1306 | * Performs an index-ascending scan of the tree for present items which | |
1307 | * have the tag indexed by @tag set. Places the items at *@results and | |
1308 | * returns the number of items which were placed at *@results. | |
1309 | */ | |
1310 | unsigned int | |
35534c86 | 1311 | radix_tree_gang_lookup_tag(const struct radix_tree_root *root, void **results, |
daff89f3 JC |
1312 | unsigned long first_index, unsigned int max_items, |
1313 | unsigned int tag) | |
1da177e4 | 1314 | { |
cebbd29e | 1315 | struct radix_tree_iter iter; |
d7b62727 | 1316 | void __rcu **slot; |
cebbd29e | 1317 | unsigned int ret = 0; |
612d6c19 | 1318 | |
cebbd29e | 1319 | if (unlikely(!max_items)) |
7cf9c2c7 NP |
1320 | return 0; |
1321 | ||
cebbd29e | 1322 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
46437f9a | 1323 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1324 | if (!results[ret]) |
1325 | continue; | |
b194d16c | 1326 | if (radix_tree_is_internal_node(results[ret])) { |
46437f9a MW |
1327 | slot = radix_tree_iter_retry(&iter); |
1328 | continue; | |
1329 | } | |
cebbd29e | 1330 | if (++ret == max_items) |
1da177e4 | 1331 | break; |
1da177e4 | 1332 | } |
7cf9c2c7 | 1333 | |
1da177e4 LT |
1334 | return ret; |
1335 | } | |
1336 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag); | |
1337 | ||
47feff2c NP |
1338 | /** |
1339 | * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a | |
1340 | * radix tree based on a tag | |
1341 | * @root: radix tree root | |
1342 | * @results: where the results of the lookup are placed | |
1343 | * @first_index: start the lookup from this key | |
1344 | * @max_items: place up to this many items at *results | |
1345 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) | |
1346 | * | |
1347 | * Performs an index-ascending scan of the tree for present items which | |
1348 | * have the tag indexed by @tag set. Places the slots at *@results and | |
1349 | * returns the number of slots which were placed at *@results. | |
1350 | */ | |
1351 | unsigned int | |
35534c86 | 1352 | radix_tree_gang_lookup_tag_slot(const struct radix_tree_root *root, |
d7b62727 | 1353 | void __rcu ***results, unsigned long first_index, |
35534c86 | 1354 | unsigned int max_items, unsigned int tag) |
47feff2c | 1355 | { |
cebbd29e | 1356 | struct radix_tree_iter iter; |
d7b62727 | 1357 | void __rcu **slot; |
cebbd29e | 1358 | unsigned int ret = 0; |
47feff2c | 1359 | |
cebbd29e | 1360 | if (unlikely(!max_items)) |
47feff2c NP |
1361 | return 0; |
1362 | ||
cebbd29e KK |
1363 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
1364 | results[ret] = slot; | |
1365 | if (++ret == max_items) | |
47feff2c | 1366 | break; |
47feff2c NP |
1367 | } |
1368 | ||
1369 | return ret; | |
1370 | } | |
1371 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot); | |
1372 | ||
0ac398ef | 1373 | static bool __radix_tree_delete(struct radix_tree_root *root, |
d7b62727 | 1374 | struct radix_tree_node *node, void __rcu **slot) |
0ac398ef | 1375 | { |
0a835c4f | 1376 | void *old = rcu_dereference_raw(*slot); |
01959dfe | 1377 | int values = xa_is_value(old) ? -1 : 0; |
0ac398ef MW |
1378 | unsigned offset = get_slot_offset(node, slot); |
1379 | int tag; | |
1380 | ||
0a835c4f MW |
1381 | if (is_idr(root)) |
1382 | node_tag_set(root, node, IDR_FREE, offset); | |
1383 | else | |
1384 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1385 | node_tag_clear(root, node, tag, offset); | |
0ac398ef | 1386 | |
01959dfe | 1387 | replace_slot(slot, NULL, node, -1, values); |
1cf56f9d | 1388 | return node && delete_node(root, node); |
0ac398ef MW |
1389 | } |
1390 | ||
1da177e4 | 1391 | /** |
0ac398ef MW |
1392 | * radix_tree_iter_delete - delete the entry at this iterator position |
1393 | * @root: radix tree root | |
1394 | * @iter: iterator state | |
1395 | * @slot: pointer to slot | |
1da177e4 | 1396 | * |
0ac398ef MW |
1397 | * Delete the entry at the position currently pointed to by the iterator. |
1398 | * This may result in the current node being freed; if it is, the iterator | |
1399 | * is advanced so that it will not reference the freed memory. This | |
1400 | * function may be called without any locking if there are no other threads | |
1401 | * which can access this tree. | |
1402 | */ | |
1403 | void radix_tree_iter_delete(struct radix_tree_root *root, | |
d7b62727 | 1404 | struct radix_tree_iter *iter, void __rcu **slot) |
0ac398ef MW |
1405 | { |
1406 | if (__radix_tree_delete(root, iter->node, slot)) | |
1407 | iter->index = iter->next_index; | |
1408 | } | |
d1b48c1e | 1409 | EXPORT_SYMBOL(radix_tree_iter_delete); |
0ac398ef MW |
1410 | |
1411 | /** | |
1412 | * radix_tree_delete_item - delete an item from a radix tree | |
1413 | * @root: radix tree root | |
1414 | * @index: index key | |
1415 | * @item: expected item | |
1da177e4 | 1416 | * |
0ac398ef | 1417 | * Remove @item at @index from the radix tree rooted at @root. |
1da177e4 | 1418 | * |
0ac398ef MW |
1419 | * Return: the deleted entry, or %NULL if it was not present |
1420 | * or the entry at the given @index was not @item. | |
1da177e4 | 1421 | */ |
53c59f26 JW |
1422 | void *radix_tree_delete_item(struct radix_tree_root *root, |
1423 | unsigned long index, void *item) | |
1da177e4 | 1424 | { |
0a835c4f | 1425 | struct radix_tree_node *node = NULL; |
7a4deea1 | 1426 | void __rcu **slot = NULL; |
139e5616 | 1427 | void *entry; |
1da177e4 | 1428 | |
139e5616 | 1429 | entry = __radix_tree_lookup(root, index, &node, &slot); |
7a4deea1 MW |
1430 | if (!slot) |
1431 | return NULL; | |
0a835c4f MW |
1432 | if (!entry && (!is_idr(root) || node_tag_get(root, node, IDR_FREE, |
1433 | get_slot_offset(node, slot)))) | |
139e5616 | 1434 | return NULL; |
1da177e4 | 1435 | |
139e5616 JW |
1436 | if (item && entry != item) |
1437 | return NULL; | |
1438 | ||
0ac398ef | 1439 | __radix_tree_delete(root, node, slot); |
612d6c19 | 1440 | |
139e5616 | 1441 | return entry; |
1da177e4 | 1442 | } |
53c59f26 JW |
1443 | EXPORT_SYMBOL(radix_tree_delete_item); |
1444 | ||
1445 | /** | |
0ac398ef MW |
1446 | * radix_tree_delete - delete an entry from a radix tree |
1447 | * @root: radix tree root | |
1448 | * @index: index key | |
53c59f26 | 1449 | * |
0ac398ef | 1450 | * Remove the entry at @index from the radix tree rooted at @root. |
53c59f26 | 1451 | * |
0ac398ef | 1452 | * Return: The deleted entry, or %NULL if it was not present. |
53c59f26 JW |
1453 | */ |
1454 | void *radix_tree_delete(struct radix_tree_root *root, unsigned long index) | |
1455 | { | |
1456 | return radix_tree_delete_item(root, index, NULL); | |
1457 | } | |
1da177e4 LT |
1458 | EXPORT_SYMBOL(radix_tree_delete); |
1459 | ||
1460 | /** | |
1461 | * radix_tree_tagged - test whether any items in the tree are tagged | |
1462 | * @root: radix tree root | |
1463 | * @tag: tag to test | |
1464 | */ | |
35534c86 | 1465 | int radix_tree_tagged(const struct radix_tree_root *root, unsigned int tag) |
1da177e4 | 1466 | { |
612d6c19 | 1467 | return root_tag_get(root, tag); |
1da177e4 LT |
1468 | } |
1469 | EXPORT_SYMBOL(radix_tree_tagged); | |
1470 | ||
0a835c4f MW |
1471 | /** |
1472 | * idr_preload - preload for idr_alloc() | |
1473 | * @gfp_mask: allocation mask to use for preloading | |
1474 | * | |
1475 | * Preallocate memory to use for the next call to idr_alloc(). This function | |
1476 | * returns with preemption disabled. It will be enabled by idr_preload_end(). | |
1477 | */ | |
1478 | void idr_preload(gfp_t gfp_mask) | |
1479 | { | |
bc9ae224 ED |
1480 | if (__radix_tree_preload(gfp_mask, IDR_PRELOAD_SIZE)) |
1481 | preempt_disable(); | |
0a835c4f MW |
1482 | } |
1483 | EXPORT_SYMBOL(idr_preload); | |
1484 | ||
460488c5 | 1485 | void __rcu **idr_get_free(struct radix_tree_root *root, |
388f79fd CM |
1486 | struct radix_tree_iter *iter, gfp_t gfp, |
1487 | unsigned long max) | |
0a835c4f MW |
1488 | { |
1489 | struct radix_tree_node *node = NULL, *child; | |
f8d5d0cc | 1490 | void __rcu **slot = (void __rcu **)&root->xa_head; |
0a835c4f | 1491 | unsigned long maxindex, start = iter->next_index; |
0a835c4f MW |
1492 | unsigned int shift, offset = 0; |
1493 | ||
1494 | grow: | |
1495 | shift = radix_tree_load_root(root, &child, &maxindex); | |
1496 | if (!radix_tree_tagged(root, IDR_FREE)) | |
1497 | start = max(start, maxindex + 1); | |
1498 | if (start > max) | |
1499 | return ERR_PTR(-ENOSPC); | |
1500 | ||
1501 | if (start > maxindex) { | |
1502 | int error = radix_tree_extend(root, gfp, start, shift); | |
1503 | if (error < 0) | |
1504 | return ERR_PTR(error); | |
1505 | shift = error; | |
f8d5d0cc | 1506 | child = rcu_dereference_raw(root->xa_head); |
0a835c4f | 1507 | } |
66ee620f MW |
1508 | if (start == 0 && shift == 0) |
1509 | shift = RADIX_TREE_MAP_SHIFT; | |
0a835c4f MW |
1510 | |
1511 | while (shift) { | |
1512 | shift -= RADIX_TREE_MAP_SHIFT; | |
1513 | if (child == NULL) { | |
1514 | /* Have to add a child node. */ | |
d58275bc MW |
1515 | child = radix_tree_node_alloc(gfp, node, root, shift, |
1516 | offset, 0, 0); | |
0a835c4f MW |
1517 | if (!child) |
1518 | return ERR_PTR(-ENOMEM); | |
1519 | all_tag_set(child, IDR_FREE); | |
1520 | rcu_assign_pointer(*slot, node_to_entry(child)); | |
1521 | if (node) | |
1522 | node->count++; | |
1523 | } else if (!radix_tree_is_internal_node(child)) | |
1524 | break; | |
1525 | ||
1526 | node = entry_to_node(child); | |
1527 | offset = radix_tree_descend(node, &child, start); | |
1528 | if (!tag_get(node, IDR_FREE, offset)) { | |
1529 | offset = radix_tree_find_next_bit(node, IDR_FREE, | |
1530 | offset + 1); | |
1531 | start = next_index(start, node, offset); | |
b7e9728f | 1532 | if (start > max || start == 0) |
0a835c4f MW |
1533 | return ERR_PTR(-ENOSPC); |
1534 | while (offset == RADIX_TREE_MAP_SIZE) { | |
1535 | offset = node->offset + 1; | |
1536 | node = node->parent; | |
1537 | if (!node) | |
1538 | goto grow; | |
1539 | shift = node->shift; | |
1540 | } | |
1541 | child = rcu_dereference_raw(node->slots[offset]); | |
1542 | } | |
1543 | slot = &node->slots[offset]; | |
1544 | } | |
1545 | ||
1546 | iter->index = start; | |
1547 | if (node) | |
1548 | iter->next_index = 1 + min(max, (start | node_maxindex(node))); | |
1549 | else | |
1550 | iter->next_index = 1; | |
1551 | iter->node = node; | |
0a835c4f MW |
1552 | set_iter_tags(iter, node, offset, IDR_FREE); |
1553 | ||
1554 | return slot; | |
1555 | } | |
1556 | ||
1557 | /** | |
1558 | * idr_destroy - release all internal memory from an IDR | |
1559 | * @idr: idr handle | |
1560 | * | |
1561 | * After this function is called, the IDR is empty, and may be reused or | |
1562 | * the data structure containing it may be freed. | |
1563 | * | |
1564 | * A typical clean-up sequence for objects stored in an idr tree will use | |
1565 | * idr_for_each() to free all objects, if necessary, then idr_destroy() to | |
1566 | * free the memory used to keep track of those objects. | |
1567 | */ | |
1568 | void idr_destroy(struct idr *idr) | |
1569 | { | |
f8d5d0cc | 1570 | struct radix_tree_node *node = rcu_dereference_raw(idr->idr_rt.xa_head); |
0a835c4f MW |
1571 | if (radix_tree_is_internal_node(node)) |
1572 | radix_tree_free_nodes(node); | |
f8d5d0cc | 1573 | idr->idr_rt.xa_head = NULL; |
0a835c4f MW |
1574 | root_tag_set(&idr->idr_rt, IDR_FREE); |
1575 | } | |
1576 | EXPORT_SYMBOL(idr_destroy); | |
1577 | ||
1da177e4 | 1578 | static void |
449dd698 | 1579 | radix_tree_node_ctor(void *arg) |
1da177e4 | 1580 | { |
449dd698 JW |
1581 | struct radix_tree_node *node = arg; |
1582 | ||
1583 | memset(node, 0, sizeof(*node)); | |
1584 | INIT_LIST_HEAD(&node->private_list); | |
1da177e4 LT |
1585 | } |
1586 | ||
d544abd5 | 1587 | static int radix_tree_cpu_dead(unsigned int cpu) |
1da177e4 | 1588 | { |
2fcd9005 MW |
1589 | struct radix_tree_preload *rtp; |
1590 | struct radix_tree_node *node; | |
1591 | ||
1592 | /* Free per-cpu pool of preloaded nodes */ | |
d544abd5 SAS |
1593 | rtp = &per_cpu(radix_tree_preloads, cpu); |
1594 | while (rtp->nr) { | |
1595 | node = rtp->nodes; | |
1293d5c5 | 1596 | rtp->nodes = node->parent; |
d544abd5 SAS |
1597 | kmem_cache_free(radix_tree_node_cachep, node); |
1598 | rtp->nr--; | |
2fcd9005 | 1599 | } |
d544abd5 | 1600 | return 0; |
1da177e4 | 1601 | } |
1da177e4 LT |
1602 | |
1603 | void __init radix_tree_init(void) | |
1604 | { | |
d544abd5 | 1605 | int ret; |
7e784422 MH |
1606 | |
1607 | BUILD_BUG_ON(RADIX_TREE_MAX_TAGS + __GFP_BITS_SHIFT > 32); | |
fa290cda | 1608 | BUILD_BUG_ON(ROOT_IS_IDR & ~GFP_ZONEMASK); |
02c02bf1 | 1609 | BUILD_BUG_ON(XA_CHUNK_SIZE > 255); |
1da177e4 LT |
1610 | radix_tree_node_cachep = kmem_cache_create("radix_tree_node", |
1611 | sizeof(struct radix_tree_node), 0, | |
488514d1 CL |
1612 | SLAB_PANIC | SLAB_RECLAIM_ACCOUNT, |
1613 | radix_tree_node_ctor); | |
d544abd5 SAS |
1614 | ret = cpuhp_setup_state_nocalls(CPUHP_RADIX_DEAD, "lib/radix:dead", |
1615 | NULL, radix_tree_cpu_dead); | |
1616 | WARN_ON(ret < 0); | |
1da177e4 | 1617 | } |